OpenAI Releases GPT-5.4 Mini and Nano — Speed Up, Price Down

OpenAI shipped two new models today: GPT-5.4 mini and GPT-5.4 nano. Both are lightweight derivatives of GPT-5.4 — the flagship model the company released on March 5 — and they extend that generation’s capabilities to users who either can’t pay or can’t afford the latency of the full model.

GPT-5.4 mini is the one that matters to most people. It’s now available on the free and Go ChatGPT tiers, which previously topped out at GPT-5.0 mini. The upgrade is substantial: OpenAI claims GPT-5.4 mini runs more than twice as fast as its predecessor, with better multimodal performance across image, audio, and web search tasks. Free users get the full GPT-5.4 generation for the first time.

GPT-5.4 nano is strictly API-only and aimed at a specific use case: subagent delegation. Think classification, entity extraction, routing decisions — tasks where a large model is overkill and latency costs compound across thousands of calls. At $0.20 per million input tokens, nano undercuts GPT-5.0 mini’s pricing by roughly 60%. That’s a meaningful signal about where OpenAI sees the agentic infrastructure market going: cheap, fast models doing the dirty work while more capable models reason at the top.

Both sit downstream of GPT-5.4’s core improvements. The parent model brought a 1-million-token context window, native computer-use, and a 33% reduction in hallucinated claims compared to GPT-5.2. The mini and nano variants inherit the improved calibration, which means fewer confident wrong answers even at the smaller scale.

The free-tier upgrade is the more commercially interesting move. OpenAI has consistently used model generation upgrades to collapse the perceived value gap between free and paid plans, then reopen it with paid-only features. This cycle — let free users catch up, then pull paid users ahead again — is working. ChatGPT’s free tier remains the most capable free AI product on the market.

For developers, the more useful data point is nano’s pricing. Subagent orchestration at scale demands cheap inference. GPT-5.4 nano at $0.20/M input tokens is competitive with Gemini 2.0 Flash and Mistral’s smallest open-source options. Whether the quality holds up at that price against open weights alternatives will determine how much of the agentic pipeline OpenAI can lock in.

The releases follow a now-familiar OpenAI cadence: ship a flagship model, then push derivative sizes out over two to three weeks to maximize market coverage. GPT-5.4 nano is the fourth model in the GPT-5.4 family in under three weeks.